Analysis of COVID in Prince George's County, Maryland¶
Author Beatrix Tran
Objective¶
The objective of this project is to gather and analyze the COVID infected and death cases in Prince George's County (PG), Maryland from March 2020 to current.
PG county is one of the biggest counties and is also the one that has the most COVID cases in Maryland. This project gathers multiple real-time statistics from the Maryland Department of Health to inform the public about the current COVID situation of PG county and the severity of the infected cities inside the county
To do this, the objective is broken down into three-part:
- Find the correlation between the infected Staff vs. Non-Staff Resident(if any)
- Analyze and search for the potential trend of the daily cases in PG
- Explore and find the correlation among some infectious area of PG (if any)
Getting start!¶
To get ready for this project you should get familiar with some python library. The information about these libraries can be found at:
- scikit-learn: for model selection regression
- pandas: for data tidying
- folium: for geographic visualization
Additional Required Libraries:¶
- requests
- json
- pandas
- re
- matplotlib.pyplot
- numpy
- datetime
Data:¶
The data for this project are gathered from the current data available at the Maryland Department of Health. For more information about the data, please visit the links below.
- MD COVID-19 Number of Deaths by Affected Congregate Facility in 7 days
- MD COVID-19 Number of Cases by Affected Congregate Facility in 7 days
- MD COVID-19 Cumulative Daily Confirmed Deaths By County
- MD COVID-19 Cumulative Daily Confirmed Cases By County
- MD COVID-19 Cumulative Daily Confirmed Cases by Zipcode
Load data¶
!pip install folium
import requests
import json
import pandas as pd
import matplotlib.pyplot as plt
#Util function to get json data from url and convert data to dataframe
def get_data(url):
json_data = requests.get(url=url) # request url
content = json.loads(json_data.content) #load data content
data = []
#extract relevant information from feature in the json structure to a list of dictionaries
for i in range(len(content['features'])):
temp = content['features'][i]
att = temp['attributes']
data.append(att)
#build a table from that dictionary
df = pd.DataFrame.from_dict(data)
return df
#get data from url
url_cases = 'https://opendata.arcgis.com/datasets/4dd7afed5b9946f282abc4d8b03ce5f8_0/FeatureServer/0/query?where=1%3D1&outFields=*&outSR=4326&f=json'
url_deaths = 'https://opendata.arcgis.com/datasets/ca12954794ce41fe9028640aa2315afd_0/FeatureServer/0/query?where=1%3D1&outFields=*&outSR=4326&f=json'
url_zipcode = 'https://services.arcgis.com/njFNhDsUCentVYJW/arcgis/rest/services/MDCOVID19_MASTER_ZIP_CODE_CASES/FeatureServer/0/query?where=1%3D1&outFields=*&outSR=4326&f=json'
url_daily_cases = 'https://services.arcgis.com/njFNhDsUCentVYJW/arcgis/rest/services/MDCOVID19_CasesByCounty/FeatureServer/0/query?where=1%3D1&outFields=*&outSR=4326&f=json'
url_daily_deaths = 'https://services.arcgis.com/njFNhDsUCentVYJW/arcgis/rest/services/MDCOVID19_ConfirmedDeathsByCounty/FeatureServer/0/query?where=1%3D1&outFields=*&outSR=4326&f=json'
url_geo_zip = 'https://opendata.arcgis.com/datasets/046ba81433b94c0d90465adec040de1a_4.geojson'
url_geo = 'https://geodata.md.gov/imap/rest/services/Boundaries/MD_PoliticalBoundaries/FeatureServer/4/query?where=1%3D1&outFields=*&outSR=4326&f=json'
#raw_cases = get_data(url=url_cases)
#raw_deaths = get_data(url=url_deaths)
raw_zip_cases = get_data(url=url_zipcode)
raw_daily_deaths = get_data(url=url_daily_deaths)
raw_daily_cases = get_data(url=url_daily_cases)
raw_geo = get_data(url=url_geo)
Simple cleaning and pre-processing data¶
Raw data sometimes was encoded or simply doesn't make sense. In this section, we will focus on making the data somewhat readable
raw_geo= raw_geo.loc[(raw_geo['ZIPCODE1'] >= '20607') & (raw_geo['ZIPCODE1'] <= '20792')]# Select only the zipcode the belong or partialy belong to PG county
raw_geo.head()
print("Raw MD COVID-19 Cases by Zipcode")
#Select only the zipcode the belong or partialy belong to PG county
raw_zip_cases = raw_zip_cases.loc[(raw_zip_cases['ZIP_CODE'] >= '20607') & (raw_zip_cases['ZIP_CODE'] <= '20792')]
raw_zip_cases = raw_zip_cases.drop(['OBJECTID'], axis=1)
raw_zip_cases = raw_zip_cases.fillna(0) #replace nan with 0
raw_zip_cases.head()
print("Raw MD COVID-19 Confirmed Deaths By County")
raw_daily_deaths = raw_daily_deaths.drop(['OBJECTID'], axis=1) #droped the OBJECTID column
raw_daily_deaths = raw_daily_deaths.fillna(0)
raw_daily_deaths.head()
print("Raw MD COVID-19 Confirmed Cases By County")
raw_daily_cases = raw_daily_cases.drop(['OBJECTID'], axis=1)
raw_daily_cases = raw_daily_cases.fillna(0)
raw_daily_cases.head()
print("Raw MD COVID-19 - Number of Deaths by Affected Congregate Facility")
raw_deaths = raw_deaths.loc[raw_deaths['COUNTY'] == "Prince George's"] #select only cases belong to PG county
raw_deaths.head()
print("Raw MD COVID-19 - Number of Deaths by Affected Congregate Facility")
raw_deaths = raw_deaths.loc[raw_deaths['COUNTY'] == "Prince George's"] #select only cases belong to PG county
raw_cases.head()
Further processing data¶
In this section, the data is further manipulated to get or create only relevant information
#Creating empty data frame
cases = pd.DataFrame()
deaths = pd.DataFrame()
zip_cases = pd.DataFrame()
In the raw COVID-19 Cases by Zipcode (raw_zip_cases), the date was used as the header for each column and was in string format that doesn't really make sense at the first glance (exp: F4_15_2020). We will use regex to reformat the date for understandability and further analysis
import re #regex
l = list(raw_zip_cases) #list of column headers
res = []
for elem in l:
temp = re.split(r'(\d.*)', elem) #split at the 1st digit (F4_15_2020 -> ['F', '4_15_2020', ''])
if len(temp) < 2:
res.append(temp[0]) #append the first element (aka Zipcode)
else:
_,date,_ = temp
date = date.replace("_","/") #re fomatting (4_15_2020 -> 4/15/2020)
res.append(date)
raw_zip_cases.columns = res #set column headers as dates
zip_cases = raw_zip_cases.drop(['ZIP_CODE'], axis=1).T.reset_index(drop=True) # transpose the table
zip_cases.columns = raw_zip_cases['ZIP_CODE'].tolist() # Use zipcode as column header
zip_cases.insert(0, 'Date', res[1:]) # created a date column
zip_cases.head()
Once the zipcode data is cleaned, we move on to the daily cases and deaths. Notice how the structure of daily cases and daily deaths are very similar. Thus, it's better to have a utility function to process both of them. This function’s purpose is to convert the Unix int64 format to datetime object and then format them to mm/dd/YYYY format for readability. Then we will apply this function to the unprocessed daily cases (raw_daily_cases) and unprocessed daily cases (raw_daily_deaths) data
#Util convert the unix int format to datetime object
import datetime
def to_dt(data):
l = []
for row in data.iterrows():
# convert to datetime and format to mm/dd/YYYY format (ex: 1584280800000 -> [2020,15,03] -> 03/15/2020)
l.append(datetime.datetime.fromtimestamp(int(row[1][0]/1000)).strftime("%m/%d/%Y"))
return l #return the list of mm/dd/YYYY
raw_daily_cases["DATE"] = to_dt(raw_daily_cases)
raw_daily_cases.head()
raw_daily_deaths["DATE"] = to_dt(raw_daily_deaths)
raw_daily_deaths.head()
Next, we want to narrow down the information to just PG county to see the trend of COVID and the correlation between cases and deaths in PG county. To do so, we create a utility function to extract the information of deaths and cases for a specific county (this case is PG).
The variables represented in these two tables are very similar. Thus, we will create another utility function to combine for later analysis
#Util to extract the cases/deaths of a particular county
def case_OR_death_per_county(raw_data, county, case_OR_death):
data = pd.DataFrame()
data["Date"] = raw_data["DATE"]#keep the date as in the original data
data[county + '_' + case_OR_death] = raw_data[county]#select only the county
return data
#Util to combine the cases/deaths of a particular county
def total_per_county(county):
daily_deaths = case_OR_death_per_county(raw_daily_deaths, county, 'Deaths')
daily_cases = case_OR_death_per_county(raw_daily_cases, county, 'Cases')
data_total_daily = pd.merge(daily_cases, daily_deaths, how="outer", on=["Date"])#using outter merge to combine both data
data_total_daily = data_total_daily.fillna(0)#convert nan to 0
return data_total_daily
print("Cumulative Cases and deaths in Prince George's county since 03/15/2020")
data_totalPG_daily = total_per_county('Prince_Georges')
data_totalPG_daily.head()
Once we get the desired cumulative cases and deaths table we can move on to the current 7 days cases and deaths table. To do so, we keep important information such as Date, Cases, Deaths. Then create two extra columns to count the number of staff and non-staff residents infected within the past 7 days
cases["Date"] = raw_cases["DATE"] #keep date
cases["Type"] = raw_cases[""] #keep type
cases["Number_of_Case(s)"] = raw_cases["Prince_George's"] #extract only cases in PG county
import numpy as np
arr = np.zeros(48, dtype=int) #create 0 array that has the same length as the cases data
ind = 0
for row in cases.iterrows():
if "Staff:" in row[1][1]: #if the word "Staff" is in the type of infected, add it to the array
arr[ind] = row[1][2]
ind = 1+ind
cases['Staff_Infected']=arr #create a colum for infected staff
cases = cases.groupby('Date', as_index=False).sum() # sum the number of infected staff by date
cases["Non-Staff_Resident"] = cases['Number_of_Case(s)']-cases['Staff_Infected'] #calculate non-staff infected cases by subtracting the total cases and the infected staff case
cases
Because the 7 days deaths data is structured different than the cases data. We will create a SUM column in the table to keep track of total deaths
raw_deaths = raw_deaths.replace('N/A', 0) #replace N/A value with 0
raw_deaths["SUM"] = raw_deaths.iloc[:,3:9].sum(axis=1) #create a sum column to count total deaths
deaths["Date"] = raw_deaths["DATE"]
deaths["Number_of_Death(s)"] = raw_deaths["SUM"]
#number of death staff is the sum of the 'Staff:_Nursing,_Assisted_Living,_Group_Homes_Affected_Facilities' and 'Staff:_State_and_Local_Affected_Facilities' column
deaths["Staff_Deaths"] = raw_deaths.loc[:,['Staff:_Nursing,_Assisted_Living,_Group_Homes_Affected_Facilities','Staff:_State_and_Local_Affected_Facilities']].sum(axis=1)
deaths = deaths.groupby('Date', as_index=False).sum() # sum the number of deaths staff by date
deaths["Non-Staff_Deaths"] = deaths['Number_of_Death(s)']-deaths['Staff_Deaths'] #calculate non-staff infected deaths by subtracting the total deaths and the deaths staff case
deaths
Now we will combine the 7 days infected cases and death to find the correlation of cases and deaths for staff and non-staff members
data_totalPG = pd.concat([cases, deaths.drop(["Date"], axis=1)], axis=1)
data_totalPG['Date'] = pd.to_datetime(data_totalPG['Date'])
data_totalPG = data_totalPG.sort_values(by='Date')
print("Cases and deaths in Prince George's county in 7 days")
data_totalPG
Evaluation¶
I. Staff vs. Non-Staff-Resident¶
The main focus of this part is to find the correlation of infection between congregate facility staff (ie. nurse, doctor, etc.) and nonstaff residents.
We will use the data_totalPG for this purpose.
Using matplotlib.pyplot, let's plot the Infected cases vs Mortality of Staff and Infected cases vs Mortality of Non-Staff Resident to see how the infection and death cases affect each group individually.
#ultil function to create bar chart to compare two or more catergories
def get_comparison_plot(df, title, x, y):
fig, ax = plt.subplots() #create subplot
plt.title(title)
df.plot.bar(x , y, ax = ax) # plot bar chart
for p in ax.patches:
ax.annotate(np.round(p.get_height(),decimals=2), (p.get_x()+p.get_width()/2., p.get_height())) # add anotation
# get comparision plot for Infected cases vs Mortality of Staff and Infected cases vs Mortality of Non-Staff Resident
get_comparison_plot(data_totalPG, "Infected cases vs Motality of Staff", x = 'Date', y = ['Staff_Deaths', 'Staff_Infected'])
get_comparison_plot(data_totalPG, "Infected cases vs Motality of Non-Staff Resident", x = 'Date', y = ['Non-Staff_Deaths', 'Non-Staff_Resident'])
From the above graphs, it seems like there isn't a correlation between death and the infected rate per group. However, there seems to be a linear relationship between the nonstaff and staff infection rate.
We now use the LinearRegression from sklearn.linear_model to verify this assumption and see how these two correlated to one another
from sklearn.linear_model import LinearRegression
from sklearn.metrics import mean_squared_error, r2_score
from sklearn.model_selection import train_test_split
#Util to fit the linear model and get coefficient, intercept, MSE, r2 score
def get_coef_intc(x, y):
x = x.values.reshape(-1,1) #make sure inpute are in correct shape
y = y.values.reshape(-1,1)
#split input into testing and training data
X_train, X_test, y_train, y_test = train_test_split(x, y, test_size=0.2, random_state=0)
#fit training model
reg1 = LinearRegression().fit(X_train, y_train)
# The coefficient
print('Coefficients: ', reg1.coef_.flatten()[0])
# The intercept
print('Intercepts: ', reg1.intercept_.flatten()[0])
# Make predictions using the testing data
y_pred = reg1.predict(X_test)
# The mean squared error
print('Mean squared error(MSE): %.2f'
% mean_squared_error(y_test, y_pred))
# The coefficient of determination: 1 is perfect prediction
print('Coefficient of determination(r2 score): %.2f'
% r2_score(y_test, y_pred))
return reg1.coef_.flatten(), reg1.intercept_.flatten()
#Util to plot the fitted line
def plot_fit(x, y, title, xla, yla):
c, i = get_coef_intc(x, y) # get coefficient and intercept
plt.title(title)
#plot data as scatter plot
plt.scatter(x, y)
plt.xlabel(xla)
plt.ylabel(yla)
#plot fited line using function y = cx + i
plt.plot(x, i + c*x, 'r', label='fitted line')
plt.show()
#Fit and Plot Infected cases Staff vs Non-Staff Resident
x = data_totalPG["Non-Staff_Resident"]
y = data_totalPG["Staff_Infected"]
title = "Infected Coefficient between Staff vs Non-Staff"
xla = "Non-Staff_Resident"
yla = "Staff_Infected"
plot_fit(x, y, title, xla, yla)
There is a positive correlation between the infection rate of infected patients and staff. The r2 score is 0.97 means that 97% of the data can be explained with this model. Their linear relationship is follow the function y = 0.42x + 43.68 . We can also say that approximately there was a staff member get infected for every 44 infected PG residents
II. Daily PG county mortality and infection rate¶
In this section, we will compare PG against its Maryland neighboring counties:
- Anne Arundel County (east)
- Calvert County (southeast)
- Charles County (south)
- Howard County (north)
- Montgomery County (northwest)
We will also attempt to try finding the infection trend and predict infection rate over time
First, we will make a daily_total table with the information on cases and deaths of PG and its neighboring counties.
print("Cumulative Cases and deaths in Montgomery county since 03/15/2020")
# Use ultil function total_per_county created from the previous part to extract cases of relevant counties
data_totalPG_daily
data_totalMont_daily = total_per_county('Montgomery')
data_totalAnne_daily = total_per_county('Anne_Arundel')
data_totalCalv_daily = total_per_county('Calvert')
data_totalChar_daily = total_per_county('Charles')
data_totalHowa_daily = total_per_county('Howard')
#combine the information
daily_total = pd.concat([data_totalPG_daily, data_totalMont_daily.drop(["Date"], axis=1),
data_totalAnne_daily.drop(["Date"], axis=1), data_totalCalv_daily.drop(["Date"], axis=1),
data_totalChar_daily.drop(["Date"], axis=1), data_totalHowa_daily.drop(["Date"], axis=1)], axis=1, sort=False)
daily_total.head()
#separate deaths and cases
l = list(daily_total.drop(["Date"], axis=1))
cs = []
ds = []
for elem in l:
if "Cases" in elem:
cs.append(elem)
else:
ds.append(elem)
ind = list(range(1, len(daily_total) + 1))
daily_total["Ind"] = ind # add an index column for plotting number of days
ax = daily_total.plot.bar(x = "Ind", y = cs, rot=0, figsize=(30,10))
ax.set_xlabel("Day number")
plt.title("Daily Cases")
dx = daily_total.plot.bar(x = "Ind", y = ds, rot=0, figsize=(30,10))
dx.set_xlabel("Day number")
plt.title("Daily Deaths")
From the plots, we can see that PG seems to have the most infected cases but not the most infected deaths. Next, let's see if we can find the model for predicting the infectious rate over time
From first glance, the relationship between infection cases and time seems to be a non-linear relationship and rather curve. We will apply polynomial regression using PolynomialFeatures provided by sklearn to test out this assumption.
from sklearn.preprocessing import PolynomialFeatures
from sklearn.metrics import mean_squared_error, r2_score
deg = 3 # polynomial degree of 3
X = daily_total['Ind'].values.reshape(-1,1) # make sure data have the correct shape
y = daily_total['Prince_Georges_Cases'].values.reshape(-1,1)
#split 80% of data for training and 20% for testing
X_train, X_test, Y_train, Y_test = train_test_split(X, y, test_size=0.2, random_state=0)
#fit training data to polynomial regression model
X_train_poly = PolynomialFeatures(degree=deg).fit_transform(X_train)
# fit the transformed features to Linear Regression
poly_model = LinearRegression().fit(X_train_poly, Y_train)
# predicting on training data
y_train_predicted = poly_model.predict(X_train_poly)
# predicting on test data
y_test_predict = poly_model.predict(PolynomialFeatures(degree=deg).fit_transform(X_test))
# evaluating the model on training dataset
mse = mean_squared_error(Y_train, y_train_predicted)
r2 = r2_score(Y_train, y_train_predicted)
# evaluating the model on test dataset
mse_test = mean_squared_error(Y_test, y_test_predict)
r2_test = r2_score(Y_test, y_test_predict)
print('Coefficients: ', poly_model.coef_.flatten())
print('Intercepts: ', poly_model.intercept_.flatten())
print("MSE of testing set is {}".format(mse_test))
print("R2 score of testing set is {}".format(r2_test))
#plot the fitted line
x = daily_total["Ind"]
y = daily_total["Prince_Georges_Cases"]
cs = poly_model.coef_.flatten()
is_ = poly_model.intercept_.flatten()
plt.figure(figsize=(30,10))
plt.ylabel('Infected')
plt.xlabel('Day')
plt.title('PG total cases')
plt.scatter(x, y)
plt.plot(x, is_[0] + cs[1]*x + cs[2]*x*x + cs[3]*x*x*x, 'r', label='fitted line')
plt.show()
There is a positive correlation between the infection rate over time. The r2 score is 0.99 means that 99% of the data can be explained with this model. We can say that the trend of cases over time is follow the function y = 646.99684683 + -1.53e+02x + 8.16e+00x^2 + -4.91e-02x^3
III. Exploring the infectous areas of PG county¶
In this section, we will explore some infectious cities in PG and see how their infectious rate affect one another.
First, we will create the map with label using mainly folium to visually compare the number of cases per city and find cities of interest.
To do so, we will create a table of cumulative infection cases per zipcode.
print("Total cases in Prince George's county related zipcode since 03/15/2020")
total_zip_cases = pd.DataFrame()
res = zip_cases.columns
total_zip_cases.insert(0, 'Zip_code', res[1:])
row = zip_cases.iloc[-1:]
row = row.drop(["Date"], axis=1).reset_index()
row = row.T[0].tolist()
total_zip_cases.insert(0, 'Total_cases', row[1:])
total_zip_cases.head()
Then we load the geojson data of MD that contain 5 digit zipcode and their boundaries coordinate. Then we store this information to a new JSON file call updated-file.json
#load geojson data as json
json_data = requests.get(url=url_geo_zip)
content = json.loads(json_data.content)
# remove ZIP codes not in our dataset
geozips = []
for i in range(len(content['features'])):
if content['features'][i]["properties"]["ZIPCODE1"] in list(raw_zip_cases['ZIP_CODE'].unique()):
geozips.append(content['features'][i])
# creating new JSON object
new_json = dict.fromkeys(['type','features'])
new_json['type'] = 'FeatureCollection'
new_json['features'] = geozips
# save JSON object as updated-file
open("updated-file.json", "w").write(
json.dumps(new_json, sort_keys=True, indent=4, separators=(',', ': '))
)
Then we generate the PG geo-map present our data with choropleth map. We use will use our previously saved JSON file to define the boundaries of each city in this choropleth map. Then we use the zip code and cases from the total_zip_cases table to label the map and set the color indicator for cases per zip code
import folium
# create a PG county geo map
m = folium.Map(location = [38.6049, -76.8721], prefer_canvas=True, zoom_start = 9)
md_geo = r'updated-file.json' # open file
choropleth = folium.Choropleth( # use Choropleth to color map of PG
geo_data=md_geo, # choose geo data
name='choropleth',
data=total_zip_cases, # choose data
columns=['Zip_code', 'Total_cases'], # use Zip_code', 'Total_cases' column of data as indecator for color
key_on='feature.properties.ZIPCODE1', # search of zipcode properties of the geo data
fill_color='PuRd',
fill_opacity=1.0,
line_opacity=0.5,
legend_name='COVID Cases in Prince Georges County ',
).add_to(m)
folium.LayerControl().add_to(m) # add layer control to map
# Display Regional Label
style_function = "font-size: 15px; font-weight: bold"
choropleth.geojson.add_child(
folium.features.GeoJsonTooltip(['ZIPName'], style=style_function, labels=False))
folium.LayerControl().add_to(m)
# save map as html
m.save(outfile = 'Feature_map.html')
m
This is the map of infectious rate of PG rank by case from 0 to 2000. The darker the color, the higer the cases.
From this map, we can see that most cases are appear in the north area of PG. Let's have a closer look of these areas
There are three places in these area that stand out the most:
College Park is the one that has the least infected cases.
Lanham is the one that has the second most infected cases.
- Hyattsville is the one that has the most infected cases.
It's interesting because these three counties are very close to one another and have very different infection rates. Let's see how their infectious rate affects one another.
To do so, we will create a new table for cases in Hyattsville, College Park, and Lanham. Then we plot the fitted line for each city vs one another and see how they are correlated.
Lanham = []
Hyattsville = []
College_Park = []
# get the zip code of each city from the raw_geo data
for row in raw_geo.iterrows():
if 'College Park' in row[1][2]:
College_Park.append(row[1][1])
if 'Hyattsville' in row[1][2]:
Hyattsville.append(row[1][1])
if 'Lanham' in row[1][2]:
Lanham.append(row[1][1])
h_cases = zip_cases[Hyattsville].sum(axis=1) #count total cases per city
l_cases = zip_cases[Lanham].sum(axis=1)
c_cases = zip_cases[College_Park].sum(axis=1)
hcl = pd.DataFrame()
hcl['Hyattsville'] = h_cases
hcl['College_Park'] = c_cases
hcl['Lanham'] = l_cases
ind = list(range(1, len(h_cases) + 1 ))
hcl.insert(0, 'Day', ind) # created a number of day column
hcl.head()
#bar plot
ax = hcl.plot.bar(x = "Day", y = list(hcl.drop(['Day'], axis=1)), rot=0, figsize=(30,10))
ax.set_xlabel("Day number")
plt.title("Hyattsville vs College_Park vs Lanham")
From the above plot, we can see clearly how the infection cases are different from one city to another. Despite this level of differences, all three cities seem to follow the trend that was generally applied for PG (which we calculated previously).
Notice that they also seem to have a linear relationship. Let's plot the linear model to then and see how one cities cases affect one another
x = hcl['Hyattsville']
y = hcl['College_Park']
title = "Will changes in Hyattsville affect College Park?"
xla = 'Hyattsville'
yla = 'College_Park'
plot_fit(x, y, title, xla, yla)
x = hcl['College_Park']
y = hcl['Hyattsville']
title = "Will changes in College Park affect Hattsville?"
xla = 'College_Park'
yla = 'Hyattsville'
plot_fit(x, y, title, xla, yla)
x = hcl['College_Park']
y = hcl['Lanham']
title = "Will changes in College Park affect Lanham?"
xla = 'College_Park'
yla = 'Lanham'
plot_fit(x, y, title, xla, yla)
x = hcl['Hyattsville']
y = hcl['Lanham']
title = "Will changes in Lanham affect Hyattsville?"
xla = 'Lanham'
yla = 'Hyattsville'
plot_fit(x, y, title, xla, yla)
x = hcl['Lanham']
y = hcl['Hyattsville']
title = "Will changes in Lanham affect Hattsville?"
xla = 'Lanham'
yla = 'Hyattsville'
plot_fit(x, y, title, xla, yla)
As expected, the high r2 score could means that our linear models can be used to explain the cases in each city. Each of them have a unique linear relationship to another that follow the equation y = their_coefficient*x + their_ntercept, where their_coefficient and their_ntercept can be found above
Conclusion and limitation¶
From the data analysis, we can say that the number of COVID patients or nonstaff residents is one of the main drivers for the COVID cases of staff in congregate facilities. To be more specific, their relationship shows that approximately there is an infected staff for every 44 infected residents. However, the data for this evaluation is very small (only 7 days). Thus, it might not be good enough to predict the future trend between staff and resident cases.
From the second part of the analysis, we can visually see how different cases in PG compare to other counties. We also predicted the trend COVID for PG with the current data set. Even though the data set is currently small, the data will expand over time and the model can become more accurate. As it is now, we can still see the trend of COVID case still increase but not as much as before. Another limitation of this data is due to the slow testing response in the beginning which might invalidate the actual number of cases in March and affect the prediction accuracy
From the third part of this analysis, we can see that the COVID trend in the city generally follows the trend of the county. How ever, the actual infected rate per city varies. This variation might due to many factors like social distancing, ages, population density, gender. For example, College Park consists of many young adults and students and also has an early response to social distancing (school closed in March, curfew started in April). This might be one of the reasons why the city has lower infected cases. More studies will need to be done to evaluate how these factors affect each city.
Message from author I hope you find this tutorial/analysis helpful. Stay safe and stay social distancing